Real-Time Self-Evolving Intelligence System with LLM Integration for Multi-Tenant Invoice Anomaly Detection
International Journal of Creative and Open Research in Engineering and Management · 29.04.2026 · 02 (04) · 1-12
DOI: 10.55041/ijcope.v2i4.889 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Creative and Open Research in Engineering and Management, online erschienen am 29.04.2026. Verfasst von Aniket Kumar Aniket Kumar und 2 weiteren Autor:innen.
Abstract (Originalsprache)
This paper presents a self-evolving, schema-adaptive anomaly detection system designed specifically for multi-tenant billing and invoicing environments. Contemporary enterprise billing pipelines suffer from persistent vulnerabilities arising from static rule-based validation frameworks that cannot adapt to the distributional evolution of transactional data over time. The proposed architecture addresses this fundamental limitation through the integration of three synergistic components: automatic schema-aware feature engineering, an incremental Isolation Forest anomaly detection engine, and a Large Language Model (LLM)-driven reasoning and explanation layer. Upon onboarding a new tenant, the system performs unsupervised schema introspection to classify data attributes and derive semantically meaningful feature representations. The anomaly detection model is subsequently trained on tenant-specific historical invoice data and undergoes periodic retraining through a human-feedback-driven incremental learning loop. When anomalous invoices are identified at runtime, the system computes feature-level deviation signatures and submits structured deviation reports to an integrated LLM, which generates contextually grounded explanations and corrective recommendations. Experimental evaluation demonstrates that this hybrid intelligence framework substantially reduces false-positive anomaly rates while improving detection sensitivity across diverse invoice schemas. The feedback-driven retraining mechanism enables sustained accuracy gains over deployment lifetime, and the LLM explanation layer measurably enhances operator trust and response effectiveness. This work constitutes a meaningful contribution toward intelligent, self-adaptive financial validation systems capable of evolving autonomously with changing business patterns and invoice structures. KEYWORDS: Anomaly Detection, Isolation Forest, Large Language Model, Invoice Validation, Schema Adaptation, Incremental Learning, Multi-Tenant Systems, Feature Engineering, Concept Drift, Intelligent Billing Systems, Human-in-the-Loop, Self-Evolving Intelligence
Zitieren
APA
Kumar, A. K. A., Kumar, P. K. P., & Singh, E. S. E. (2026). Real-Time Self-Evolving Intelligence System with LLM Integration for Multi-Tenant Invoice Anomaly Detection. International Journal of Creative and Open Research in Engineering and Management, 02(04), 1-12. https://doi.org/10.55041/ijcope.v2i4.889
Vancouver
Kumar AKA, Kumar PKP, Singh ESE. Real-Time Self-Evolving Intelligence System with LLM Integration for Multi-Tenant Invoice Anomaly Detection. International Journal of Creative and Open Research in Engineering and Management. 2026;02(04):1-12. doi:10.55041/ijcope.v2i4.889
BibTeX
@article{Kumar2026,
author = {Kumar, Aniket Kumar Aniket and Kumar, Pawan Kumar Pawan and Singh, Eklavya Singh Eklavya},
title = {Real-Time Self-Evolving Intelligence System with LLM Integration for Multi-Tenant Invoice Anomaly Detection},
journal = {International Journal of Creative and Open Research in Engineering and Management},
year = {2026},
volume = {02},
number = {04},
pages = {1--12},
doi = {10.55041/ijcope.v2i4.889}
}
EUVIMED